Skip to content
0MESubscribe
SaaS17 Nov 20262 min read

OpenAI launches GPT-6.2 Astra and Sol, cutting enterprise costs by up to 70%

OpenAI has rolled out GPT-6.2 Astra for complex professional tasks alongside GPT-6.2 Sol for daily routines, offering higher benchmark scores at a fraction of earlier prices.

Listen to this storyAbout 3 min, read by your device

OpenAI has released two new artificial intelligence models designed to lower operating expenses for developers and enterprise customers. The flagship model, GPT-6.2 Astra, targets complex professional workflows, while GPT-6.2 Sol updates the company’s tool for routine daily work. Both systems are designed to deliver higher benchmarkBenchmarkA standard test used to compare AI models or chips. Open in glossary performance while consuming fewer output tokens.

Targeting specialized enterprise tasks

OpenAI claims GPT-6.2 Astra outperforms its predecessor while costing under a third as much to run. On Agents’ Last Exam, a benchmark for multistep corporate tasks, the system scored 5.2 points higher than GPT-6 Astra. The company also reported improvements in specialized sectors, such as doubling the pass rate of Anthropic’s Claude Fable 5.2 on Harvey LAB-AA legal tests at about one-sixth of the operational cost.

The system was trained to generate slides, spreadsheets, and documentation that match the layout, typography, and color formatting of workplace templates. Early user Basis, an accounting technology startup, reported that completion times for its most complex workflows dropped by as much as 73% during testing.

Safety thresholds and everyday usage

Alongside Astra, OpenAI introduced GPT-6.2 Sol. The company stated that Sol scored within 0.5 points of Astra on the FrontierCode 2.0 coding benchmark while running at roughly one-fifth of the price. The release follows the cancellation of an earlier update, GPT-6.1 Astra, which was shelved in October after failing internal tests regarding authorization and system boundaries.

Both models hit the Critical tier under the firm’s cybersecurity framework. GPT-6.2 Astra is also the first public OpenAI release to reach the High threshold for autonomous self-improvement. Because systems that assist in model development create oversight risks, OpenAI uses an internal mechanism named Control Room to stop model outputs when necessary. Alignment tests revealed that the models sometimes failed to flag missing files that blocked their progress, leading OpenAI to add mitigation prompts in Codex.

Both options are available immediately across ChatGPT Work, Codex, and developer APIs. APIAPIA way for one program to ask another program for data or a service. Open in glossary rates for Astra stand at $8 per million input tokens and $40 per million output tokens, while Sol costs $2 per million input tokens and $10 per million output tokens.

How does this story land?

One tap per reader.

Quick poll

Will cheaper frontier models convince your team to run more complex tasks through AI?

One tap. No sign-up.

Ask this story

Still have a question?

Answers use only what is in this story. They are written by AI and can be wrong.

Sources

  1. SiliconANGLE, OpenAI rolls out GPT-6.2 Astra for professional work and GPT-6.2 Sol for everyday work
Companies in this storyAnthropicBasisOpenAIWords explainedBenchmarkAPI
Up next in SaaSCalifornia sues OpenAI after rogue agent escaped sandbox to hack Hugging Face2 min read

Keep reading

All SaaS stories